Update app.py

app.py

@@ -14,6 +14,86 @@ from constraint_parser_llm import parse_constraints
 from constraint_parser import parse_constraints
 from semantic_ranker import score_courses  # optional; keep if you want semantic boost
 
+# -------- Robust free-text constraint parsing (deterministic fallback) --------
+import re
+
+DAY_ALIASES = {
+    "sunday": "Sun", "sundays": "Sun", "sun": "Sun",
+    "monday": "Mon", "mondays": "Mon", "mon": "Mon",
+    "tuesday": "Tue", "tuesdays": "Tue", "tue": "Tue",
+    "wednesday": "Wed", "wednesdays": "Wed", "wed": "Wed",
+    "thursday": "Thu", "thursdays": "Thu", "thu": "Thu",
+    "friday": "Fri", "fridays": "Fri", "fri": "Fri",
+    "saturday": "Sat", "saturdays": "Sat", "sat": "Sat",
+}
+WEEKEND = {"Sat", "Sun"}
+WEEKDAYS = {"Mon", "Tue", "Wed", "Thu", "Fri"}
+
+def _norm_ampm(h: str, m: str|None, ap: str|None) -> str:
+    h_i = int(h)
+    m_s = (m or "00")
+    ap = (ap or "").upper()
+    if ap not in ("AM", "PM"):
+        # heuristic: <= 8 -> AM, >= 4 -> PM; otherwise assume AM
+        ap = "AM" if h_i <= 11 else "PM"
+    return f"{h_i}:{m_s} {ap}"
+
+def parse_det_constraints(text: str):
+    """
+    Deterministic, high-recall extraction of:
+      - banned_days: {"Mon","Tue",...}
+      - no_before: "H:MM AM/PM"  (start times >= this)
+      - no_after:  "H:MM AM/PM"  (end times   <= this)
+    """
+    res = {"banned_days": set(), "no_before": None, "no_after": None}
+    if not text:
+        return res
+    t = text.lower().strip()
+
+    # Weekends / weekdays
+    if re.search(r"\bno (?:weekend|weekends)\b", t) or re.search(r"\bweekdays only\b|\bonly on weekdays\b", t):
+        res["banned_days"] |= WEEKEND
+    if re.search(r"\bweekends only\b|\bonly on weekends\b", t) or re.search(r"\bno weekdays\b", t):
+        res["banned_days"] |= WEEKDAYS
+
+    # Ban specific days with broad triggers
+    triggers = r"(?:no|avoid|except|skip|without|not on|exclude|ban|block|never on)"
+    for alias, abbr in DAY_ALIASES.items():
+        # e.g., "no class on monday", "avoid mon", "not on tuesdays"
+        if re.search(rf"\b{triggers}\s+(?:classes?|class|lectures?)?\s*(?:on\s*)?{re.escape(alias)}\b", t):
+            res["banned_days"].add(abbr)
+
+    # Time windows: "only start after 10 AM", "after 10", "not before 10", "no earlier than 10"
+    m_after = re.search(r"\b(?:only\s*)?(?:start\s*)?after\s+(\d{1,2})(?::(\d{2}))?\s*(am|pm)?\b", t)
+    m_not_before = re.search(r"\b(?:not before|no earlier than)\s+(\d{1,2})(?::(\d{2}))?\s*(am|pm)?\b", t)
+    if m_after or m_not_before:
+        h, mm, ap = (m_after or m_not_before).groups()
+        res["no_before"] = _norm_ampm(h, mm, ap)
+
+    # "before 6 PM", "end before 6", "not after 6", "no later than 6"
+    m_before = re.search(r"\b(?:before|end before|not after|no later than)\s+(\d{1,2})(?::(\d{2}))?\s*(am|pm)?\b", t)
+    if m_before:
+        h, mm, ap = m_before.groups()
+        res["no_after"] = _norm_ampm(h, mm, ap)
+
+    # "mornings / afternoons / evenings" shortcuts
+    if re.search(r"\bno mornings?\b", t):
+        res["no_after"] = "12:00 PM"
+    if re.search(r"\bmornings?\b", t) and not re.search(r"\bno mornings?\b", t):
+        res["no_before"] = res["no_before"] or "10:00 AM"
+
+    if re.search(r"\bafternoons?\b", t):
+        res["no_before"] = res["no_before"] or "12:00 PM"
+    if re.search(r"\bno afternoons?\b", t):
+        res["no_after"] = "12:00 PM"
+
+    if re.search(r"\bevenings?\b", t):
+        res["no_before"] = res["no_before"] or "4:00 PM"
+    if re.search(r"\bno evenings?\b", t):
+        res["no_after"] = "4:00 PM"
+
+    return res
+
 # Robust day parsing for phrases like "no class on monday", "avoid mon", "not on mondays", etc.
 DAY_ALIASES = {
     "sunday": "Sun", "sundays": "Sun", "sun": "Sun",